See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318749834

Automatic Speech Recognition: A Review

Article  in  International Journal of Computer Applications · December 2012

DOI: 10.5120/9722-4190

CITATIONS READS

11 4,044

2 authors, including:

Rishi pal Singh

Guru Jambheshwar University of Science & Technology
48 PUBLICATIONS   145 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Rishi pal Singh on 06 September 2017.

The user has requested enhancement of the downloaded file.


Automatic Speech Recognition: A Review
Shipra J. Arora
Research Scholar, CSE Department
GJUST, Hisar
ABSTRACT
This paper attempts to describe a literature review of Automatic
Speech Recognition. It discusses past years advances made so
as to provide progress that has been accomplished in this area of
research. One of the important challenges for researchers is
ASR accuracy. The Speech recognition System focuses on
difficulties with ASR, basic building blocks of speech
processing, feature extraction, speech recognition and
performance evaluation. The main objective of the review paper
is to bring to light the progress made for ASRs of different
languages and the technological viewpoint of ASR in different
countries and to compare and contrast the techniques used in
various stages of Speech recognition and identify research topic
in this challenging field. We are not presenting exhaustive
descriptions of systems or mathematical formulations but
rather, we are presenting distinctive and novel features of
selected systems and their relative merits and demerits.
Keywords
Automatic speech recognition, Language Model, Speech
Processing, Database, Pattern Recognition, Hidden Markov
Model.
1. INTRODUCTION
We ask that authors The Speech is one of the most important
tools for communication between human and his environment.
Therefore manufacturing of ASR is need for human being all
the time. Speech recognition made it feasible for machine to
understand human languages. As information technology has a
bang on more and more aspects of our lives with every year, the
problem of communication between human beings and
information processing devices becomes increasingly
significant. Up to now, communication has almost fully been
through the use of keyboards and screens, but speech is the
most widely used, natural and the fastest means of
communication for people. In a speech recognition system,
many parameters affect the accuracy of the Recognition
System. These parameters are: dependence or independence
from speaker, discrete or continuous word recognition,
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
Rishi Pal Singh
CSE Department
GJUST, Hisar
vocabulary, environment, acoustic model, language model,
perplexity, transducer etc. Problems such as noisy
environment, different pronouncing of one word by one person
in several times, dissimilar expressing of one word by two
different speakers, incompatibility between train and test
conditions led to made system without complete recognition.
Resolving each of these problems is a good step towards this
aim.
2. SPEECH AND LANGUAGE
PROCESSING
2.1 Basic Building Blocks
Figure 1 shows some building blocks that perform
transformations pertinent to speech recognition. A waveform
modifier Figure 1(a) takes an input speech signal and produces
a modified signal. The modification might be a clipping of large
values of the signal; a frequency spectrum filtering that alters
the shape of signal or enhances the speech and de-emphasises
noise that is present. A symbol transducer Figure 1(b) can take
in one discrete symbol sequence and yields a modified sequence
on its output. If the input were a sequence of words in one
language and the output were an equivalent word sequence in
another language, this transducer would be a language
translation device. Parameter extractor Figure 1(c) takes an
input speech signal and yielding parameters of speech wave. In
recognizers, it is often called the pre-processor. A feature
extractor Figure 1(d) can receive parameters and produce a
more abstract set of important information carrying features
such as determining what portions of speech are voiced,
whether the sound is loud and resonant like a vowel etc. A
segmenter and labeller Figure 1(e) can receive the set of
features and produce a linear string of phonemes or other
identified segments. The unit identifier Figure 1(f) takes input
symbol sequence which may be compared to the expected
reference sequences for various units to determine what
linguistic units appear to be in the input. The most common unit
identifier is a ‘word matcher’ which finds the closest matching
word, based on which word’s stored pronunciation string is
most like the input string. With these building blocks, we have
the essential prerequisites for discussing the main knowledge
sources needed for machine understanding of speech.
34

Figure 1 Basic Building Blocks for processing speech and language
2.2 Types of Speech
Speech can be classified into following categories
2.2.1 Isolated words:
Isolated word recognisers accept single word at a time. These
systems have "Listen/Not-Listen" states, where they require
the speaker to wait between utterances. Isolated Utterance
might be a better name for this class.
2.2.2 Connected words
Connected word speech recognition is the system where the
words are separated by pauses. Connected word speech
recognition is a class of fluent speech strings where the set of
strings is derived from small-to-moderate size vocabulary
such as digit strings, spelled letter sequences, combination of
alphanumeric. Like isolated word speech recognition, this set
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
too has a property that the basic speech-recognition unit is the
word/phrase to much extent.
2.2.3 Continuous speech:
Continuous speech recognition deals with the speech where
words are connected together instead of being separated by
pauses. As a result unknown boundary information about
words, co-articulation, production of surrounding phonemes
and rate of speech effect the performance of continuous
speech recognition systems. Recognizers with continuous
speech capabilities are some of the most difficult to create
because they utilize special methods to determine utterance
boundaries.
2.3 Classification of ASR system
Classification of speech recognition system is shown in Figure
2.
35

Speech Processing
Analysis/Synthesis Speech Recognition
Speaker Recognition Speech Recognition
Speech Mode Speaker Mode Vocabulary Size
Isolated Speech Speaker Independent
Continuous Speech Speaker dependent
Figure 2 Speech Processing Classification
Speech recognition systems are classified as discrete or
continuous systems that are speaker dependent or independent.
Discrete systems maintain a separate acoustic model for each
word, combination of words or phrases referred to as isolated
word speech recognition (ISR). Continuous speech recognition
(CSR) systems responds to a user who pronounces words,
phrases or sentences that are in a series of specific order. A
speaker-dependent system requires that the user record an
example of the word, sentence or phrase prior to its being
recognized by the system. i.e the user trains the system. A
speaker-independent system does not require any recording
prior to system use. It is developed to operate for any speaker of
a particular type. Speaker-dependent systems are simpler to
construct and are more accurate than speaker-independent
systems. As a result, the focus of early voice recognition
systems was primarily speaker-dependent isolated word
systems that used limited vocabulary. At the time, overcoming
the restrictions in the state of technology required a greater
focus on human-to-computer interaction. The challenge was to
identify how improved speech recognition technology could be
used to support the enhancement of human interaction with
machines. An important element in the creation of speech
recognition system is the size of the vocabulary. Vocabulary
affects the complexity and accuracy of the system. Size of the
vocabulary can be small, medium or large. Obviously, it is
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
Coding
Language Identification
Speaking Style
Small Dictation
Medium Spontaneous
Large
much easier to look up one of 50 words in 50-word dictionary
rather than one of the hundreds of thousands of words in a
Webster’s dictionary. Another important qualifier in the
determination of the complexity of a speech recognition
system is the type of speech that the recognition system uses:
discrete or continuous. In a discrete speech system, the user
must pause between each word which makes speech recognition
task much easier. Continuous speech is more difficult because
of several reasons. First, it is difficult to find the start and end
boundary of words. Another problem is that the production of
surrounding phonemes affect the production of each phoneme.
Also speech rate affect the recognition of continuous speech.
2.4 Difficulties with ASR
Following are the some of the difficulties with ASR
2.4.1 Human comprehension of speech
Human use the knowledge about the speaker and the subject
while listening more than the ears. Words are not sequences
together arbitrarily but there is a grammatical structure that
human use to predict words not yet uttered. In ASR, we have
only speech signal. We can construct a model for grammatical
structure and use some statistical model to improve prediction
but there are still the problem how to model world knowledge
36

and the knowledge of speaker. Of course, we can not model
world knowledge but the question is how much we actually
require to measure up to human comprehension in the ASR.
2.4.2 Spoken language is not equal to written
language
Spoken language is less critical than written language and
human make much more performance error while speaking.
Speech is two-way communication as compared to written
communication which is one-way. The most important issue is
disfluencies in speech e.g. normal speech contains repetitions,
tongue slips, change of subject in the middle of an utterance etc.
In the last few years, it has become clear that spoken language
is different from written language. In ASR, we have to identify
and address these differences.
2.4.3 Noise
Speech is uttered in an environment of sound such as ticking a
clock, another speaker in the background, TV playing in
another room etc. This unwanted information in the speech
signal is known as noise. In ASR, we have to identify these
noise and filter out it from the speech signal. Echo effect is
another kind of noise in which speech signal is bounced on
some surrounding object and it appears in the microphone a few
milliseconds later.
2.4.4 Body Language
A human speaker not only communicates with speech but also
with body gestures such as waving hands, moving eyes,
postures etc. In ASR, such information is not available. This
problem is addressed in Multimodality research area where
studies are conducted to incorporate body language to improve
human-machine communication.
2.4.5 Channel Variability
Variability is the context in which acoustic wave is uttered.
Different types of microphones, noise that changes over time
and anything that affects the content of acoustic wave from the
speaker to the discrete representation in a computer is known as
channel variability.
2.4.6 Speaker Variability
Human speak differently. The voice is not only different
between speakers but also wide variation within one specific
speaker. Some of the variations are given below
2.4.6.1 Speaking Style
All speakers speak differently due to their unique personality.
They have a distinctive way to pronounce words. Speaking
style varies in different situations. We do not speak in the same
way in the public area, with our teachers or friends. Humans
also express emotions while speaking i.e happy, sad, fear,
surprise or anger.
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
2.4.6.2 Realization
If same word were pronounced again and again, the resulting
speech signal would never be same. There will be some small
differences in the acoustic wave produced. Speech realization
changes over time.
2.4.6.3 Speaker Sex
Male and Female have different voices. Female have shorter
vocal tract than male. That is why the pitch of female voice is
roughly twice than male.
2.4.6.4 Dialects
Dialects are group related variations with in a language.
Regional dialect: It involves features of vocabulary,
pronunciation and grammar according to geographical area the
speaker belongs.
Social dialect: It involves features of vocabulary, pronunciation
and grammar according to social group of speaker.
In many cases, we may be forced to consider dialects as another
language because of the large differences between two dialects.
We have considered some of the difficulties of speech
recognition but not all. The most problematic issue is strong
variability. Our goal is efficient user interface not natural verbal
communication.
3 SPEECH RECOGNITION
TECHNIQUES CLASSIFICATION
Basically there are three approaches to speech recognition.
 Acoustic Phonetic Approach
 Pattern Recognition Approach
 Artificial Intelligence Approach
3.1 Acoustic Phonetic Approach
Acoustic phonetic approach (Hemdal and Hughes 1967),
postulates that there exist distinguishing phonetic units in
spoken language and these units are characterized by a set of
acoustics properties. The acoustic properties of phonetic units
are highly variable both with speakers and with neighbouring
sounds i.e co-articulation effect, it is understood in this
approach that the rules governing the variability are
straightforward and can be readily learned by a machine. The
first stage in this approach is the speech spectral analysis .The
next stage is a feature extraction stage that converts the spectral
measurements into a set of features that describe the acoustic
properties of the different phonetic units. The next stage is a
segmentation and labelling stage in which the speech signal is
segmented into isolated regions, followed by attaching one or
more phonetic labels to each segmented region. The last stage
finds a valid word from the phonetic label sequences produced
by the segmentation to labelling. The acoustic phonetic
approach has not been widely used .
37
 International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012

Speech
Recognition
Techniques

Pattern Artificial
Acoustic Phonetic
Recogniton Intelligence
Approach
Approach Approach

Model Based Rule Based

Approach Approach

Template Based Knowledge Based

Approach Systems

HMM,SVM,DTW,
Bayesian Classification
Approach

Figure 3 Speech Recognition Techniques Classification

3.2 Pattern Recognition Approach
The pattern-matching approach (Rabiner and Juang 1993)
shown in Figure 4 is used to extract patterns based on certain
conditions and to separate one class from the others. .It has
four stages i.e. feature measurement, pattern training , pattern
classification and decision logic. In order to define a test
pattern, a sequence of measurements is made on the input
signal. Reference pattern is created by taking one or more test
patterns corresponding to speech sounds. This can be in the
form of a speech template or a statistical model i.e.HMM and
can be applied to a sound , a word, or a phrase. In the pattern
classification stage of the approach, a direct comparison is
made between the unknown test pattern and class reference
pattern and a measure of distance between them is computed.
Decision logic stage determines the identity of the unknown
according to the integrity of match of the patterns. This
approach has become the prime method for speech
recognition in the last six decades . A block diagram of this
approach is shown below. There exists two methods i.e.
template method and stochastic model method.
3.2.1 Template Method
In this method, unknown speech is compared against a set of
templates in order to find the best match. During the last six
decades, template based method to speech recognition have
provided a family of techniques. A collection of prototypical
speech patterns are stored as reference patterns representing
the dictionary of candidate’s words. Recognition is then
carried out by matching an unknown spoken utterance with
each of these reference templates and selecting the category of
the best matching pattern. This method provides good
recognition performance for a variety of practical
applications. But it has the disadvantage that variations in
speech can be modelled by using many templates per word,
which eventually becomes unfeasible.

38
 International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012

Pattern Templates
Training or models

Analysis Test
System
Reference Patterns

S(n) pattern

Recognized Speech
Pattern
Decision
Classifier
Logic

Filter Bank/LPC Local Distance Measure

Dynamic Time Warping

Figure 4 Pattern Recognition Approach

Training System :

Signal Si COMPUTER SYSTEM IN

TRAINING MODE
Declared message Identity
(or intended response)

Storage of one or more STORED TEMPLATES FOR

Representative Pronunciation MESSAGE CATEGORIES

Recognition System :

COMPUTER SYSTEM IN
Signal Sx RECOGNITION MODE Selection of the
closet of the
reference patterns
STORED TEMPLATES FOR
MESSAGE CATEGORIES

Figure 5 Template matching Method

39

3.2.2 Stochastic method
Stochastic methods depict the use of probabilistic models to
deal with incomplete information. In speech recognition,
incompleteness arises from various sources. e.g. speaker
variability, contextual effect etc. The most popular stochastic
approach now a day is hidden Markov modelling (HMM). A
HMM is characterized by a finite state Markov model and a
set of output distributions. The transition parameters in the
Markov models are temporal variability’s, while the
parameters in the output distribution model are spectral
variability’s. These two types of variabilites are the core of
speech recognition. It is more general as compared to template
based approach. The primary problems for HMM design are
a) The evaluation of the probability of a sequence of
observations given a specific HMM. b) The determination of a
best sequence of modal states and c) the adjustment of modal
parameters so as to best account for the observed signal. Once
these fundamental problems are solved, we can apply HMMs
to selected problems in speech recognition.
3.3 Artificial intelligence approach
The Artificial Intelligence approach is a fusion of the acoustic
phonetic approach and pattern recognition approach. Some
researchers developed recognition system by using acoustic
phonetic knowledge to develop classification rules for speech
sounds. While template based methods provided little insight
about human speech processing, but these methods have been
very effective in the design of a variety of speech recognition
systems. On the other hand, linguistic and phonetic literature
provided insights about human speech processing. However,
this approach had only partial success due to the
complicatedness in quantifying expert knowledge. Another
problem is the integration of levels of human knowledge i.e.
phonetics, lexical access, syntax, semantics and pragmatics.
4 PHASES OF ASR
Automatic speech recognition system involves two phases:
Training phase and recognition phase. A rigorous training
procedure is followed to map the basic speech unit such as
phone, syllable to the acoustic observation. In training phase,
known speech is recorded, pre-processed and then enters the
first stage i.e. Feature extraction. The next three stages are
HMM creation, HMM training and HMM storage. The
recognition phase starts with the acoustic analysis of unknown
speech signal. The signal captured is converted to a series of
acoustic feature vectors. Using appropriate algorithm, the
input observations are processed. The speech is compared
against the HMM‟s networks and the word which is
pronounced is displayed. An ASR system can only recognize
what it has learned during the training process. But, the
system is able to recognize even those words, which are not
present in the training corpus and for which sub-word units of
the new word are known to the system and the new word
exists in the system dictionary.
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
4.1 Modules of ASR
Modules for a speech recognition system are shown in Figure
6.
i. Speech Signal acquisition
ii. Feature Extraction
iii. Acoustic Modelling
iv. Language & Lexical Modelling
v. Recognition
Two of these modules Speech acquisition and Feature
extraction are common to both the phases of ASR.
Figure 6 ASR block diagram
Model adaptation is meant for minimizing the dependencies
on speakers‟ voice, acoustic environment, microphones and
transmission channel, and to improve the generalization
capability of the system.
4.1.1 Feature Extraction
Feature extraction requires much attention because
recognition performance relies heavily on the feature
extraction phase. Different techniques for feature extraction
are LPC, MFCC, AMFCC, RAS, DAS, ΔMFCC, Higher lag
autocorrelation coefficients, PLP, MF-PLP, BFCC, RPLP etc.
It has been found that noise robust spectral estimation is
possible on the higher lag autocorrelation coefficients.
Therefore, eliminating the lower lags of the noisy speech
signal autocorrelation leads to removal of the main noise
components.
4.1.2 Acoustic Model
Acoustic model is the main component for an ASR which
accounts for most of the computational load and performance
of the system. The Acoustic model is developed for detecting
the spoken phoneme. Its creation involves the use of audio
recordings of speech and their text scripts and then compiling
them into a statistical representation of sounds which make up
words.
4.1.3 Lexical Models
To provide the pronunciation of each word in a given
language, Lexicon is developed. Various combinations of
phones are defined through lexicon model to give valid words
40

for the recognition. Neural networks have helped to develop
4.1.4 Language Models
Language model is the main component operated on million
of words, consisting of millions of parameters and with the
help of pronunciation dictionary ,developed word sequences
in a sentence. ASR systems uses n-gram language models
which are used to search for correct word sequence by
predicting the likelihood of the nth word on the basis of the
n−1 preceding words. The probability of occurrence of a word
sequence W is calculated as:
P (W) =P (w1, w2, . . . . . . . . . . . . . ., w m-1, wm) =P(w1).
P(w2|w1). P(w3|w1w2)............................. P(w m |w1w2w3
................w m-1).
For large vocabulary speech recognizers, two problems occur
during the construction of n-gram language models. For real
applications, large amount of training data generally leads to
large models. Second is the sparseness problem, is being
faced during the training of domain specific models.
Language models are non-deterministic. Both these features
make it complicated.
5 LITERATURE SURVEY
Speech recognition came into existence during 1920. The first
machine i.e. Radio Rex ,a toy to recognize voice was
manufactured. Bell Labs developed a speech synthesis
machine at the World fair in New York. But later on they
discarded efforts based on an incorrect conclusion that the AI
is ultimately required for success. In order to develop systems
for ASR, attempts were made in 1950s where researchers
studied the fundamental concepts of phonetic-acoustic. Most
of the systems in 1950[1] for recognizing speech examines the
vowels spectral resonancews of each utterances. At Bell Labs
Davis, Biddulph and Balashek(1952) premeditated a isolated
digit recognition system for a single speaker[2] using formant
frequencies estimated during vowel regions of each digit. At
RCA Labs, Olson and Belar (1950) built 10 syllables
recognizer of a single speaker[3] and Forgie and Forgie built a
speaker-independent 10-vowel recognizer [4] at MIT Lincoln
Lab, by measuring spectral resonances for vowels. Fry and
Denes(1959) tried to build a phoneme recognizer to recognize
four vowels and nine consonants [5] at University College in
England by using a spectrum analyser and a pattern matcher
to make the recognition decision. Japanese labs entering
recognition field in 1960-70. As computers are not fast
enough, they designed special purpose H/W as a part of their
system. In Tokyo, Nagata et.al described a system of the
Radio Research lab, was a H/W vowel recognizer[6]. Another
effort was the work of Sakai and Doshita in 1962, of Kyoto
University who built a H/W phoneme recognizer[7]. In 1963,
Nagata and co-workers at NEC Labs built a the digit
recognizer[7]. This led to a long productive research program.
In 1970 , the key focus of research was on isolated word
recognition. IBM researchers studied in large vocabulary
speech recognition. At AT&T Bell Labs, researchers began
speaker independent speech recognition experiments[8]. A
large number of clustering algorithms were used to find the
number of distinct patterns required to represent words to
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
lexical model for non-native speech recognition.
achieve speaker independent speech recognition. This
research has been refined so that the techniques for speaker
independent patterns are widely used. Carnegie Mellon
University’s Harphy system[9] recognize speech with
vocabulary size of 1011 words with reasonable accuracy. It
was the first to make use of finite state network to reduce
computation and determine the closest matching strings
efficiently.
In 1980, the key focus of research was on connected words
speech recognition. In the beginning of 1980, Moshey J. Lasry
studied speech spectrogram of letters and digits[10] and
developed a feature based speech recognition. There is a
change in technology in 1980 from template based approaches
to statistical modelling approach specially HMM [11,12] in
speech research. The most significant paradigm shift has been
the introduction of statistical methods, especially stochastic
processing with HMM( Baker, 1975 & Jelinek, 1976) in the
early 1970’s (Portiz 1988). More than 30 years later, this
methodology still predominates. Despite their simplicity, N-
gram language models have proved remarkably powerful.
Now days, most practical speech recognition systems are
based on statistical approach and their results with additional
improvements have been made in 1990s. In 1980, Hidden
Markov model(HMM) approach is one of the key
technologies developed. IBM, Institute for Defence Analysis
(IDA) and Dragon Systems understood HMM , but it was not
renowned in the mid-1980s. Neural networks to speech
recognition problems is the another technology that was
reintroduced in the late 1980s.
In 1990, Pattern recognition approach was developed. It
followed Bayes’s framework traditionally but it has been
altered into an optimization problem with minimization of the
empirical recognition error[13]. The reason for this alteration
is that the distribution functions for the speech signal could
not be chosen accurately and under these conditions, Bayes’
theory can not be applied. However, aim is to design
recognizer with least recognition error rather than best fitting
to given data. The techniques used for error minimization are
Minimum Classification error (MCE) and Maximum Mutual
Information(MMI). These techniques led to maximum
likelihood based approach[14] to speech recognition
performance. A weighted HMM algorithm is proposed to
address HMM based speech recognition issues of robustness
and discrimination,. In order to decrease the acoustic
mismatch between given set of speech model and test
utterance, a maximum likelihood stochastic matching
approach[15] was proposed. A narrative approach [16] for
HMM speech recognition system is based on the use of a
Neural network as a vector quantizer which is remarkable
innovation in training the neural network. Nam Soo Kim et.al.
described a variety of methods for estimating a robust output
probability distribution based on HMM [17]. An extension of
the viterbi algorithm[18] made second order HMM efficient
as compared to existing viterbi algorithm. In 1990s, the
DARPA program continued. After that the centre of attention
is Air Travel Information Service (ATIS) task and later focus
on transcription of broadcast news (BN). Advances in
41
 International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012

continuous speech recognition and noisy environment speech
recognition, have been explained. In the area of noisy robust
speech recognition, minor work has been done.For noisy
environment, for robust speech recognition, a new approach
to an auditory model was proposed[26]. This approach is
computationally efficient as compared with other models. A
model based spectral estimation algorithm has been
developed[27].
In 2000, a variational Bayesian estimation technique was
developed[19]. It is based on posterior distribution of
parameters. Giuseppe Richardi have developed a technique to
solve the problem of adaptive learning[20] in ASR. In 2005,
some improvements have been made on Large Vocabulary
Continuous Speech recognition system for performance
improvement[21]. A 5-year national project Corpus of
Spontaneous Japanese (CSJ) [22] was conducted in Japan. It
consists of 7 millions of words approximately, corresponding
to speech of 700 hours. The techniques used in this project
are acoustic modelling, sentence boundary detection,
pronunciation modelling, acoustic as well as language model
adaptation, and automatic speech summarization [23].
Utterance verification is being investigated [24] to further
increase the robustness of speech recognition systems,
especially for spontaneous speech,. When humans speak to
each other, they use multimodal communication. It increases
the rate of successful transfer of information when the
communication take place in a noisy environment. In speech
recognition, the use of the visual face information, specially
lip movement, has been examined, and results show that using
both mode of information gives better performances than
using only the audio or only the visual information, specially,
in noisy environment.
SCARF: It is a software toolkit designed for doing speech
recognition with the help of segmental conditional random
fields.
MICROPHONES: They are being used by researchers for
recording speech database. Sony and I-ball has developed
some microphones which are unidirectional and noiseless.
7 PERFORMANCE OF SPEECH
RECOGNITION SYSTEM
The performance of speech recognition is specified in terms
of accuracy and speed. Accuracy is measured in terms of
performance accuracy which is known as word error rate
(WER) whereas speed is measured with the real time factor.
Word Error Rate
It is a common metric of the speech recognition performance.
As recognized word sequence have a different length from the
reference word sequence, there is difficulty in measuring
performance.
S DI
WER = N
where S is number of substitutions
D is number of deletions
I is number of insertions
and N is number of words in the reference.
Sometimes word recognition rate (WRR) is used instead of
WER while describing performance of speech recognition.
WRR = 1- WER

6 TOOLS FOR ASR S DI

Following are the various tools used for ASR N
= 1-

PRAAT: It is free software with latest version 5.3.04 which

N S DI
can run on wide range of OS platforms and meant for = N
recording and analysis of human speech in mono or stereo
Speed
AUDACITY: It is free, open source software available with It is measured by real time factor. If it takes time T to process
latest version of 1.3.14(Beta) which can run on wide range of an input of duration D then real time factor is defined by
OS platforms and meant for recording and editing sounds.
T
CSL: Computerised Speech Lab is a highly advanced speech D
RTF =
and signal processing workstation (software and hardware). It
possesses robust hardware for data acquisition and a versatile RTF ≤1 implies real time processing.
suite of software for speech analysis.
8 SUMMARY
HTK: The basic application of open source Hidden Markov Research in speech recognition has been carried out
Toolkit (HTK), written completely in ANSI C, is to build and intensively for the last 60 years. The technological progress
manipulate hidden Markov models. can be summarized in Table-1[28]

SPHINX: Sphinx 4 is a latest version of Sphinx series of

speech recognizer tools, written completely in Java
programming language. It provides a more flexible framework
for research in speech recognition.

42
 International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012

Table 1. SPEECH RECOGNITION SUMMARIZATION 11 REFERENCES

[1]. Sadaoki Furui, November 2005, 50 years of Progress in
Sr.No. Past Present speech and Speaker Recognition Research , ECTI
Transactions on Computer and Information
Template Corpus-based statistical Technology,Vol.1. No.2.
1
matching modelling,e.g.HMM, n-gram [2]. K.H.Davis, R.Biddulph, and S.Balashek, 1952,
grams Automatic Recognition of spoken Digits,
2 Distance–based Likelihood –based methods J.Acoust.Soc.Am.,24(6):637-642.
methods
[3]. H.F.Olson and H.Belar, 1956, Phonetic Typewriter ,
3 Maximum Discriminative approach e.g. J.Acoust.Soc.Am.,28(6):1072-1081.
likelihood MCE/GPD and MMI
Approach [4]. J.W.Forgie and C.D.Forgie, 1959, Results obtained from
4 Isolated word Continuous speech a vowel recognition computer program ,
recognition recognition J.Acoust.Soc.Am., 31(11),pp.1480-1489.

5 Small Vocabulary
6 Clean speech
recognition
7 Single speaker
recognition
8 Read speech
recognition
Large Vocabulary [5] D.B.Fry, 1959, Theoritical Aspects of Mechanical speech
Recognition , and P.Denes, The design and Operation of
the Mechanical Speech Recognizer at Universtiy College
Noisy/Telephone speech London, J.British Inst. Radio Engr., 19:4,211-299.
recognition
[6]. K.Nagata, Y.Kato, and S.Chiba, 1963, Spoken Digit
Speaker- Recognizer for Japanese Language , NEC Res.Develop.,
independent/adaptive No.6.
recognition [7]. T.Sakai and S.Doshita, 1962 The phonetic typewriter,
Spontaneous speech
recognition information processing 1962 , Proc.IFIP Congress.

Multimodal speech [8]. L.R.Rabiner, S.E.Levinson, A.E.Rosenberg, and

9 Single modality
recognition
9 DISCUSSIONS AND CONCLUSIONS
Speech is the most prominent and primary mode of
communication between human beings. Over the past five
decades, research in the area of speech is a first step towards
ordinary man-machine communication. We also have
encountered some limitations. What we know about speech
processing is very limited. This paper attempts to give a
comprehensive survey of research in speech recognition and
some year-wise progress to this date and its current status.
Although significant amount of work has been done in the last
two decades but there is still work to be done.
At present research is focussing on creating and developing
systems that would be much more robust against variability
and shift in acoustic environment, speaker characteristics,
language characteristics, external noise sources etc. It has
been found that HMM is the best technique in developing
language model. Speech recognition is very fascinating
problem. It has attracted scientists and researchers and created
a technological bang on society and is expected to boom
further in this area of man machine interaction.
10 ACKNOWLEDGEMENT
I would like to thank my husband and son for their
unwavering support. My special thanks to Dr. S.S.Agrawal,
Dr. R.K.Jain, Dr. Harsh Vardhan Kamrah and Mrs. Neelima
Kamrah for their kind support and encouragement.
J.G.Wilpon, August 1979, Speaker Independent
Recognition of Isolated Words Using Clustering
Techniques , IEEE Trans. Acoustics, Speech, Signal
Proc., ASSP-27:336-349.
[9]. B.Lowrre, 1990, The HARPY speech understanding
system ,Trends in Speech Recognition, W.Lea,Ed.,
Speech Science Pub., pp.576-586.
[10]. R.K.Moore, 1994, Twenty things we still don t know
about speech , Proc. CRIM/ FORWISS Workshop on
Progress and Prospects of speech Research an
Technology.
[11]. J.Ferguson, 1980, Hidden Markov Models for Speech,
IDA,Princeton, NJ.
[12].L.R.Rabiner, February 1989, A Tutorial on Hidden
Markov Models and Selected Applications in Speech
Recognition , Proc.IEEE,77(2):257-286.
[13].B.H.Juang and S.Furui, 2000, Automatic speech
recognition and understanding: A first step toward
natural human machine communication ,
Proc.IEEE,88,8,pp.1142-1165.
[14].K.P.Li and G.W.Hughes, 1974, Talker differences as
they appear in correlation matrices of continuous speech
spectra , J.Acoust.Soc.Am. , 55,pp.833-837.
[15] Ananth Sankar, May 1996, “ A maximum likelihood
approach to stochastic matching for Robust Speech
recognition”, IEEE Transactions on Audio, Speech and
Language processing Vol.4,No.3.
[16].Gerhard Rigoll, Jan.1994, “Maximum Mutual
Information Neural Networks for Hybrid connectionist-
HMM speech Recognition Systems “, IEEE Transactions
on Audio, Speech and Language processing Vol.2,No.1,
PartII.

43

[17].Nam Soo Kim et.al., July1995, On estimating Robust
probability Distribution in HMM in HMM based speech
recognition , IEEE Transactions on Audio, Speech and
Language processing Vol.3,No.4.
[18].Jean Francois, Jan.1997, Automatic Word Recognition
Based on Second Order Hidden Markov Models , IEEE
Transactions on Audio, Speech and Language processing
Vol.5,No.1.
[19].Mohamed Afify and Olivier Siohan, January 2004,
Sequential Estimation With Optimal Forgetting for
Robust Speech Recognition , IEEE Transactions On
Speech And Audio Processing, Vol. 12, No. 1.
[20].Giuseppe Riccardi, July 2005, Active Learning: Theory
and Applications to Automatic Speech Recognition ,
IEEE Transactions On Speech And Audio Processing,
Vol. 13, No. 4.
[21]. Mohamed Afify, Feng Liu, Hui Jiang, July 2005, A New
Verification-Based Fast-Match for Large Vocabulary
Continuous Speech Recognition , IEEE Transactions On
Speech And Audio Processing, Vol. 13, No. 4.
[22].S. Furui, 2005, Recent progress in corpus-based
spontaneous speech recognition, IEICE Trans. Inf. &
Syst., E88-D, 3, pp. 366-375.
View publication stats
International Journal of Computer Applications (0975 – 8887)
Volume 60– No.9, December 2012
[23].S. Furui, 2004, Speech-to-text and speech-to-speech
summarization of spontaneous speech, IEEE Trans.
Speech & Audio Processing, 12, 4, pp. 401-408.
[24].Eduardo Lleida et.al. March 2000, Utterance
Verification In Decoding And Training Procedures ,
IEEE Transactions On Speech And Audio Processing,
Vol. 8, No. 2.
[25] Geoff Bristow, 1986, “Electronic Speech recognition:
Techniques, Technology and Applications’’ , Collins .
[26] Doh-Suk Kim, 1999, “ Auditory processing of Speech
Signals for Robust Speech Recognition in Real World
Noisy Environment”, IEEE Transactions on Speech and
Audio Processing Vol.7,No.1.
[27] Adoram Erell et.al. ,1993, “ Filter bank energy estimation
using mixture and Markov models for Recognition of
Noisy Speech” IEEE Transactions on Audio, Speech and
Language processing Vol.1,No.1.
[28] M.A.Anusuya, S.K.Katti, 2009, “Speech Recognition by
Machine: A Review”, International Journal of Computer
Science and Information Security, vol. 6, No. 3.
44